Re: ppc64 ABI

[Rich Felker <dalias@libc.org>]

On Sun, Apr 19, 2015 at 03:40:20PM +0100, Justin Cormack wrote:
> On 19 April 2015 at 14:54, Szabolcs Nagy <nsz@port70.net> wrote:
> > * Justin Cormack <justin@specialbusservice.com> [2015-04-19 12:44:15 +0100]:
> >> I am looking at doing a ppc64 port, and after some discussion on irc
> >> last night wondering which ABI to port to.
> >>
> >> ppc64 has two ABIs:
> >> v1: as documented here
> >> v2, usually known as powerpc64le, helpfully. It is not little endian
> > ...
> >>
> >> Both have 16 byte long double oddness.
> >
> > printf/scanf decimal conversion and libm are broken with ibm128
> >
> > if the toolchain could be configured to use ieee128 then that works
> > (i see -mabi=ieeelongdouble compiler option in gcc, it seems 32bit
> > powerpc sysv abi used to require ieee128 long double, but that was
> > not implemented for aix and darwin so i'm not sure who uses it)
> >
> > 64bit long double works too and that's what musl uses on 32bit powerpc
> >
> > (i dont know how much code breaks if we change long double:
> > i think on typical desktop/server it does not matter, but if you
> > want to run some ppc optimized fortran code then it might..)
> 
> The Fedora ppc64 gcc accepts -mlong-double-64 it turns out.

OK. For a native toolchain we'd want to be able to make this the
default, but if there's no immediate easier way, it could be put in a
specfile for now.

Rich